From 98f3ba8550c198e15b2f59da70e8b2bb2a766dbe Mon Sep 17 00:00:00 2001 From: Paul Garlick Date: Wed, 30 Mar 2016 21:09:09 +0100 Subject: pyfrm2xdmf: include third, fourth, fifth and sixth order elements --- pyfrm2xdmf | 259 ++++++++++++++++++++++++++++++++++++------------------------- 1 file changed, 154 insertions(+), 105 deletions(-) (limited to 'pyfrm2xdmf') diff --git a/pyfrm2xdmf b/pyfrm2xdmf index 93dd9d3..1a47fd6 100755 --- a/pyfrm2xdmf +++ b/pyfrm2xdmf @@ -13,7 +13,8 @@ def meshFile(param): global base base, ext = os.path.splitext(param) if ext.lower() != '.pyfrm': - raise argparse.ArgumentTypeError('Mesh file must have a .pyfrm extension') + raise argparse.ArgumentTypeError('Mesh file must have a .pyfrm ' + 'extension') if not os.path.exists(param): raise argparse.ArgumentTypeError("{} not found".format(param)) return param @@ -22,13 +23,15 @@ def writeHeader(xdmfFile): "write XDMF header" xdmfFile.write('\n') xdmfFile.write('\n') - xdmfFile.write('\n') + xdmfFile.write('\n') xdmfFile.write(' \n') return def writeTopology(xdmfFile, tType, nCells, connFile): "write Topology element" - xdmfFile.write(' \n'.format(tType, nCells)) + xdmfFile.write(' \n'.format(nCells)) xdmfFile.write(' \n'.format(connFile)) xdmfFile.write(' \n') return @@ -36,14 +39,22 @@ def writeTopology(xdmfFile, tType, nCells, connFile): def writeGeometry(xdmfFile, nDims, nCells, nodeArray, pyfrm, dataset): "write Geometry element" nVerts = len(nodeArray) + # co-ordinates in separate arrays if nDims == 2: - xdmfFile.write(' \n') # co-ordinates in separate arrays + xdmfFile.write(' \n') else: - xdmfFile.write(' \n') # co-ordinates in separate arrays + xdmfFile.write(' \n') + for coord in range(nDims): - xdmfFile.write(' \n'.format(' ; '.join("$" + str(k) for k in range(nVerts)))) - writeHyperSlab(xdmfFile, coord, nDims, nCells, nodeArray, pyfrm, dataset) + xdmfFile.write(' \n'.format( + ' ; '.join( "$" + str(k) for k in range(nVerts)))) + writeHyperSlab( + xdmfFile, + coord, nDims, nCells, + nodeArray, pyfrm, dataset) xdmfFile.write(' \n') xdmfFile.write(' \n') return @@ -55,16 +66,21 @@ def writeHyperSlab(xdmfFile, coord, nDims, nCells, nodeArray, pyfrm, dataset): xdmfFile.write(' \n') - xdmfFile.write(' \n') - xdmfFile.write(' {:<3} 0 {}\n'.format(vert, coord)) # select vertex and co-ordinate (format is vertex, cell, co-ordinate) - xdmfFile.write(' 1 1 1\n') # select every cell, for this vertex and co-ordinate - xdmfFile.write(' 1 {} 1\n'.format(nCells)) # loop over cells + # select vertex and co-ordinate (format is vertex, cell, co-ordinate) + xdmfFile.write(' {:<3} 0 {}\n'.format(vert, coord)) + # select every cell, for this vertex and co-ordinate + xdmfFile.write(' 1 1 1\n') + # loop over cells + xdmfFile.write(' 1 {} 1\n'.format(nCells)) xdmfFile.write(' \n') xdmfFile.write(' \n') xdmfFile.write(' {}:/{}\n'.format(pyfrm, dataset)) xdmfFile.write(' \n') @@ -77,7 +93,8 @@ def writeAttribute(xdmfFile, tag): xdmfFile.write(' \n') - xdmfFile.write(' {}\n'.format(tag)) # tag with partition number + # tag with partition number + xdmfFile.write(' {}\n'.format(tag)) xdmfFile.write(' \n') xdmfFile.write(' \n') return @@ -88,13 +105,13 @@ def writeConnectivities(connFile, nCells, nVerts): cf.write('\n') - + for i in range (0, nCells): cf.write(' ') for j in range(nVerts): cf.write(' ' + repr(j*nCells+i).ljust(1)) cf.write('\n') - + cf.write('\n') cf.close() print('connectivities written to ' + connFile) @@ -108,21 +125,29 @@ def writeFooter(xdmfFile): def readH5lsOutput(os_output): "read output from 'h5ls' command" - tet4, tet10, tet20, tet35, tet56, tet84 = ({} for i in range(6)) # tet cells - pri6, pri18, pri40, pri75, pri126, pri196 = ({} for i in range(6)) # pri cells - pyr5, pyr14, pyr30, pyr55, pyr91, pyr140 = ({} for i in range(6)) # pyr cells - hex8, hex27, hex64, hex125, hex216 = ({} for i in range(5)) # hex cells - tri3, tri6, tri10, tri15, tri21 = ({} for i in range(5)) # tri cells - quad4, quad9, quad16, quad25, quad36 = ({} for i in range(5)) # quad cells + # tetrahedral cells (first order to sixth order) + tet4, tet10, tet20, tet35, tet56, tet84 = ({} for i in range(6)) + # prism cells + pri6, pri18, pri40, pri75, pri126, pri196 = ({} for i in range(6)) + # pyramid cells + pyr5, pyr14, pyr30, pyr55, pyr91, pyr140 = ({} for i in range(6)) + # hexahedral cells (first order to fifth order) + hex8, hex27, hex64, hex125, hex216 = ({} for i in range(5)) + # triangular cells + tri3, tri6, tri10, tri15, tri21 = ({} for i in range(5)) + # quadrilateral cells + quad4, quad9, quad16, quad25, quad36 = ({} for i in range(5)) for line in os_output.splitlines(): - spt = re.search('spt', line.decode()) # restrict to 'spt' arrays + # restrict to 'spt' arrays + spt = re.search('spt', line.decode()) if spt: chunk = line.decode().split() partno = int(re.search('\d+', chunk[0]).group()) nnodes = int(re.search('(\d+),', line.decode()).group(1)) ncells = int(re.search(' (\d+),', line.decode()).group(1)) - if re.search('quad', line.decode()): # check whether cells are quadrilaterals + # check whether cells are quadrilaterals + if re.search('quad', line.decode()): if nnodes == 4: quad4[partno] = ncells elif nnodes == 9: @@ -135,7 +160,8 @@ def readH5lsOutput(os_output): quad36[partno] = ncells else: print("unknown cell order") - elif re.search('tri', line.decode()): # check whether cells are triangles + # check whether cells are triangles + elif re.search('tri', line.decode()): if nnodes == 3: tri3[partno] = ncells elif nnodes == 6: @@ -148,7 +174,8 @@ def readH5lsOutput(os_output): tri21[partno] = ncells else: print("unknown cell order") - elif re.search('hex', line.decode()): # check whether cells are hexahedrons + # check whether cells are hexahedrons + elif re.search('hex', line.decode()): if nnodes == 8: hex8[partno] = ncells elif nnodes == 27: @@ -161,7 +188,8 @@ def readH5lsOutput(os_output): hex216[partno] = ncells else: print("unknown cell order") - elif re.search('pyr', line.decode()): # check whether cells are pyramids + # check whether cells are pyramids + elif re.search('pyr', line.decode()): if nnodes == 5: pyr5[partno] = ncells elif nnodes == 14: @@ -176,7 +204,8 @@ def readH5lsOutput(os_output): pyr140[partno] = ncells else: print("unknown cell order") - elif re.search('pri', line.decode()): # check whether cells are prisms + # check whether cells are prisms + elif re.search('pri', line.decode()): if nnodes == 6: pri6[partno] = ncells elif nnodes == 18: @@ -191,7 +220,8 @@ def readH5lsOutput(os_output): pri196[partno] = ncells else: print("unknown cell order") - elif re.search('tet', line.decode()): # check whether cells are tetrahedrons + # check whether cells are tetrahedrons + elif re.search('tet', line.decode()): if nnodes == 4: tet4[partno] = ncells elif nnodes == 10: @@ -210,27 +240,26 @@ def readH5lsOutput(os_output): print("unknown cell type") break - return (tet4, tet10, tet20, tet35, tet56, tet84, - pri6, pri18, pri40, pri75, pri126, pri196, - pyr5, pyr14, pyr30, pyr55, pyr91, pyr140, - hex8, hex27, hex64, hex125, hex216, - tri3, tri6, tri10, tri15, tri21, - quad4, quad9, quad16, quad25, quad36) + # list of cell dictionaries + cellDictList = [tet4, tet10, tet20, tet35, tet56, tet84, + pri6, pri18, pri40, pri75, pri126, pri196, + pyr5, pyr14, pyr30, pyr55, pyr91, pyr140, + hex8, hex27, hex64, hex125, hex216, + tri3, tri6, tri10, tri15, tri21, + quad4, quad9, quad16, quad25, quad36] + + return cellDictList # read command line arguments -parser = argparse.ArgumentParser(description="extract connectivities from mesh file") +parser = argparse.ArgumentParser( + description="extract connectivities from mesh file" + ": write xdmf file") parser.add_argument("mesh", help="mesh file (.pyfrm)", type=meshFile) args = parser.parse_args() # use 'h5ls' command to provide array dimensions h5lsOutput = check_output(["h5ls", args.mesh]) - -(ntet4, ntet10, ntet20, ntet35, ntet56, ntet84, -npri6, npri18, npri40, npri75, npri126, npri196, -npyr5, npyr14, npyr30, npyr55, npyr91, npyr140, -nhex8, nhex27, nhex64, nhex125, nhex216, -ntri3, ntri6, ntri10, ntri15, ntri21, -nquad4, nquad9, nquad16, nquad25, nquad36) = readH5lsOutput(h5lsOutput) +partitions = readH5lsOutput(h5lsOutput) # cell types firstOrderCellType = ['tet', 'tet', 'tet', 'tet', 'tet', 'tet', @@ -239,87 +268,107 @@ firstOrderCellType = ['tet', 'tet', 'tet', 'tet', 'tet', 'tet', 'hex', 'hex', 'hex', 'hex', 'hex', 'tri', 'tri', 'tri', 'tri', 'tri', 'quad', 'quad', 'quad', 'quad', 'quad'] -ndims = {'quad': 2, 'tri': 2, 'hex': 3, 'pyr': 3, 'pri': 3, 'tet':3} -xdmfTopologyType = {'quad': 'Quadrilateral', 'tri': 'Triangle', 'hex': 'Hexahedron', - 'pyr' : 'Pyramid', 'pri': 'Wedge', 'tet': 'Tetrahedron'} +xdmfTopologyType = {'quad': 'Quadrilateral', 'tri': 'Triangle', + 'hex' : 'Hexahedron', 'pyr': 'Pyramid', + 'pri' : 'Wedge', 'tet': 'Tetrahedron'} +ndims = {'quad': 2, 'tri': 2, + 'hex' : 3, 'pyr': 3, 'pri': 3, 'tet': 3} -# node identification (see pyfr/readers/nodemaps.py): reduces high-order cells to first order -# example, for second-order pyramid: +# node identification: reduces high-order cells to first order +# (see pyfr/readers/nodemaps.py) +# Example, for second-order pyramids having 14 solution points and 5 vertices: # >>> from pyfr.readers.nodemaps import GmshNodeMaps # >>> [GmshNodeMaps.to_pyfr['pyr', 14][i] for i in range(5)] nodeIDs = {} -nodeIDs[29] = [0, 2, 8, 6] # quad9 -nodeIDs[28] = [0, 1, 3, 2] # quad4 -nodeIDs[24] = [0, 2, 5] # tri6 -nodeIDs[23] = [0, 1, 2] # tri3 -nodeIDs[19] = [0, 2, 8, 6, 18, 20, 26, 24] # hex27 -nodeIDs[18] = [0, 1, 3, 2, 4, 5, 7, 6] # hex8 -nodeIDs[13] = [0, 2, 8, 6, 13] # pyr14 -nodeIDs[12] = [0, 1, 3, 2, 4] # pyr5 -nodeIDs[7] = [0, 2, 5, 12, 14, 17] # pri18 -nodeIDs[6] = [0, 1, 2, 3, 4, 5] # pri6 -nodeIDs[1] = [0, 2, 5, 9] # tet10 -nodeIDs[0] = [0, 1, 2, 3] # tet4 +nodeIDs[0] = [0, 1, 2, 3] # tet4 +nodeIDs[1] = [0, 2, 5, 9] # tet10 +nodeIDs[2] = [0, 3, 9, 19] # tet20 +nodeIDs[3] = [0, 4, 14, 34] # tet35 +nodeIDs[4] = [0, 5, 20, 55] # tet56 +nodeIDs[5] = [0, 6, 27, 83] # tet84 +nodeIDs[6] = [0, 1, 2, 3, 4, 5] # pri6 +nodeIDs[7] = [0, 2, 5, 12, 14, 17] # pri18 +nodeIDs[8] = [0, 3, 9, 30, 33, 39] # pri40 +nodeIDs[9] = [0, 4, 14, 60, 64, 74] # pri75 +nodeIDs[10] = [0, 5, 20, 105, 110, 125] # pri126 +nodeIDs[11] = [0, 6, 27, 168, 174, 195] # pri196 +nodeIDs[12] = [0, 1, 3, 2, 4] # pyr5 +nodeIDs[13] = [0, 2, 8, 6, 13] # pyr14 +nodeIDs[14] = [0, 3, 15, 12, 29] # pyr30 +nodeIDs[15] = [0, 4, 24, 20, 54] # pyr55 +nodeIDs[16] = [0, 5, 35, 30, 90] # pyr91 +nodeIDs[17] = [0, 6, 48, 42, 139] # pyr140 +nodeIDs[18] = [0, 1, 3, 2, 4, 5, 7, 6] # hex8 +nodeIDs[19] = [0, 2, 8, 6, 18, 20, 26, 24] # hex27 +nodeIDs[20] = [0, 3, 15, 12, 48, 51, 63, 60] # hex64 +nodeIDs[21] = [0, 4, 24, 20, 100, 104, 124, 120] # hex125 +nodeIDs[22] = [0, 5, 35, 30, 180, 185, 215, 210] # hex216 +nodeIDs[23] = [0, 1, 2] # tri3 +nodeIDs[24] = [0, 2, 5] # tri6 +nodeIDs[25] = [0, 3, 9] # tri10 +nodeIDs[26] = [0, 4, 14] # tri15 +nodeIDs[27] = [0, 5, 20] # tri21 +nodeIDs[28] = [0, 1, 3, 2] # quad4 +nodeIDs[29] = [0, 2, 8, 6] # quad9 +nodeIDs[30] = [0, 3, 15, 12] # quad16 +nodeIDs[31] = [0, 4, 24, 20] # quad25 +nodeIDs[32] = [0, 5, 35, 30] # quad36 + +# number of supported cell types +numCellTypes = len(partitions) # keys are partition numbers -tet4Keys = list(ntet4.keys()) -tet10Keys = list(ntet10.keys()) -pri6Keys = list(npri6.keys()) -pri18Keys = list(npri18.keys()) -pyr5Keys = list(npyr5.keys()) -pyr14Keys = list(npyr14.keys()) -hex8Keys = list(nhex8.keys()) -hex27Keys = list(nhex27.keys()) -tri3Keys = list(ntri3.keys()) -tri6Keys = list(ntri6.keys()) -quad4Keys = list(nquad4.keys()) -quad9Keys = list(nquad9.keys()) -# concatenate keys -allKeys = (tet4Keys + tet10Keys + - pri6Keys + pri18Keys + - pyr5Keys + pyr14Keys + - hex8Keys + hex27Keys + - tri3Keys + tri6Keys + - quad4Keys + quad9Keys) +allKeys = [] +for cd in partitions: + allKeys += list(cd.keys()) + # number of types present in each partition -numTypes = Counter(allKeys) +numTypes = Counter(allKeys) + # list of partition keys -partKeys = list(numTypes.keys()) -# list of cell dictionaries -partitions = [ntet4, ntet10, ntet20, ntet35, ntet56, ntet84, - npri6, npri18, npri40, npri75, npri126, npri196, - npyr5, npyr14, npyr30, npyr55, npyr91, npyr140, - nhex8, nhex27, nhex64, nhex125, nhex216, - ntri3, ntri6, ntri10, ntri15, ntri21, - nquad4, nquad9, nquad16, nquad25, nquad36] -numCellTypes = len(partitions) +partKeys = list(numTypes.keys()) if partKeys: # write files g = open(os.path.join(base + '.xdmf'), 'w') writeHeader(g) - + for part in partKeys: - g.write(' \n'.format(part)) + g.write( + ' \n'.format(part)) + # step through all supported cell types for cellType in range(numCellTypes): - if part in partitions[cellType]: # check whether these cells exist in this partition - xfname = os.path.join('con_' + firstOrderCellType[cellType] + '_p' + str(part) + '.xml') - dname = os.path.join('spt_' + firstOrderCellType[cellType] + '_p' + str(part)) - g.write(' \n'.format(xdmfTopologyType[firstOrderCellType[cellType]], part)) - writeTopology(g, xdmfTopologyType[firstOrderCellType[cellType]], - partitions[cellType][part], xfname) - writeGeometry(g, ndims[firstOrderCellType[cellType]], - partitions[cellType][part], - nodeIDs[cellType], - args.mesh, dname) + # check whether these cells exist in this partition + if part in partitions[cellType]: + idname = firstOrderCellType[cellType] + '_p' + str(part) + xfname = os.path.join('con_' + idname + '.xml') + dname = os.path.join('spt_' + idname) + g.write( + ' \n'.format( + xdmfTopologyType[firstOrderCellType[cellType]], + part)) + writeTopology( + g, + xdmfTopologyType[firstOrderCellType[cellType]], + partitions[cellType][part], + xfname) + writeGeometry( + g, + ndims[firstOrderCellType[cellType]], + partitions[cellType][part], + nodeIDs[cellType], + args.mesh, + dname) writeAttribute(g, part) g.write(' \n') # connectivities file - writeConnectivities(xfname, partitions[cellType][part], - len(nodeIDs[cellType])) + writeConnectivities( + xfname, + partitions[cellType][part], + len(nodeIDs[cellType])) g.write(' \n') - + writeFooter(g) g.close() else: - print("mesh does not contain any supported cell types") + print("no supported cell types") -- cgit